Image Rhumb Protocol

Open protocol for AI workflow

Rhumb Protocol

AI sessions produce work that evaporates. Plans live in chat, context vanishes between conversations, and handoffs lose everything that matters. Rhumb Workflow Protocol™ gives that work a durable spine: plain-file plans, state, handoffs, and audits that survive across tools, agents, and sessions.

Constant-bearing workflow artifacts that stay portable, inspectable, and auditable across tools.

Vendor-neutral Plain-file Protocol-first
RWP 0.31.0

Why it exists

AI workflow needs a protocol layer, not another private project format.

RWP™ standardizes the artifacts around complex work so a plan can move between people, agents, vendors, and implementations without losing the audit trail.
01

Plain files, not a platform lock-in

RWP™ artifacts are Markdown, YAML, JSON Schema, and templates. Teams can adopt them with any editor, CI system, or AI assistant.

02

Handoffs become durable context

Plans, state, manifests, and handoffs preserve what was requested, what changed, and what the next session needs to know.

03

Reference implementation, not vendor dependency

YAKKL® Meridian™ implements Rhumb™, but the protocol is independent, Apache-2.0, and designed for third-party implementations.

Artifact model

Five core artifacts keep complex work navigable.

Start with the minimum and add structure as the workflow grows. The protocol is advisory by design: RWP recommends; it does not enforce.
Markdown

PLAN.md

Defines phases, deliverables, dependencies, and verification.

YAML

INTAKE.yaml

Captures requirements, constraints, pain points, and success criteria.

YAML

manifest.yaml

Registers files, outputs, prompts, handoffs, and audit artifacts.

YAML

state.yaml

Tracks execution status, current phase, claims, heartbeats, and history.

YAML

HO-*.yaml

Preserves phase transition context between agents, people, and sessions.

Protocol layer

The missing standard between AI tools and durable work.

LSP standardized code intelligence. MCP standardized tool integration. RWP standardizes workflow memory: the plans, state, handoffs, and audits that keep AI-assisted work portable and inspectable.
For tool vendors

Implement the artifact contract, not a competing format

JSON Schema validation, conformance levels, and implementation profiles let tools adopt RWP incrementally without runtime dependencies or vendor lock-in.

For AI providers

Your users' work persists between sessions

RWP gives AI-assisted work a durable spine without requiring your platform to own the format. The protocol complements your tool by making every session's output reusable.

For developers

Plain files you can read, version, and validate

Markdown plans, YAML state, JSON Schema validation, and ABNF grammar for phase notation. No proprietary format, no service dependency, no magic.

Quick start

Adopt the protocol before adopting a tool.

The first useful version is small: one intake, one plan, one state file, and a clean handoff when the work crosses a boundary.
  1. Capture the request in INTAKE.yaml.
  2. Break the work into phases in PLAN.md.
  3. Track execution in state.yaml.
  4. Write handoffs at every phase boundary.
  5. Validate against schemas as adoption matures.

Try it now

mkdir -p rwp/plans/MP-0001 rwp/plans/MP-0001/handoffs
cat > rwp/plans/MP-0001/PLAN.md <<'EOF'
---
id: MP-0001
title: My first RWP plan
status: planning
---
# MP-0001: My first RWP plan
## Phase Breakdown
### P-01: Initial implementation
- [ ] Define deliverables
- [ ] Execute and verify
EOF

Papers

Deep dives for executives, tool builders, and implementers.

Three papers explain the protocol at different depths. Read them on the site or download the PDFs.

Integrations

Adapters for current AI tools, with Meridian as the reference implementation.

Rhumb is implementable anywhere. The current repository includes adapters for CLI and browser AI assistants, while Meridian is the reference product implementation.